Understanding coefficients of the Spatial Filter boxes

Concerning processing components: filters, file load/save, visualizations, communication ...
Post Reply
hackr
Posts: 12
Joined: Fri Aug 25, 2017 1:41 am

Understanding coefficients of the Spatial Filter boxes

Post by hackr »

Hi,

I am new to OpenViBE and am trying to understand classification. I'm currently working with the example scenario classification-0-training.xml.

My understanding thusfar is the noise generator was used to simulate data you'd receive from an EEG device via the Acquisition server/client (or a recording, etc). From there, the Simple DSP box is applied to amplify the signal with a simple transformation (though I don't understand the logic behind the example equation).

Next there are the Spatial Filter boxes. This is where I get confused. The example says that there are 4 input and 4 output channels per box, but there are a lot more coefficients. What do the coefficients correspond to? How do you know how many there should be?

I'm also confused by the fact that they are hard-coded. I'm used to the concept of coefficients being in the context of a statistical model, wherein the coefficients are determined by the data (rather than hard-coded) and each coefficient corresponds to one feature of the data.

Finally, I'm confused by the text box in the example that says some coefficients are set to .95, which results in imperfect/overlapping classification, whereas the classification is perfect if they are set to 0. If that's the case why would they be set to .95? How was it determined that 0 is ideal for those coefficients - trial and error? :?:
Last edited by hackr on Thu Aug 31, 2017 6:07 pm, edited 1 time in total.

jtlindgren
Posts: 775
Joined: Tue Dec 04, 2012 3:53 pm
Location: INRIA Rennes, FRANCE

Re: Understanding coefficients of the Spatial Filter boxes

Post by jtlindgren »

Hi Hackr,

the scenario is intended to illustrate how classifiers work in simple conditions.

The point of spatial filter is to mix the sources. What is done in that scenario is to implement a simple generative model to make data that is used to train/test the classifier part. Since we generate the data, we are of course allowed to specify any mechanism for it we like. The scenario in question has a relatively simple generation so its hopefully understandable what sort of data the classifier is receiving.

If x is a input sample vector, the spatial filter implements the matrix vector product Ax, with A the coefficient matrix.

To see how classifiers are used in BCI, please see scenarios in the bci-examples/ folder.

Happy hacking,
Jussi

hackr
Posts: 12
Joined: Fri Aug 25, 2017 1:41 am

Re: Understanding coefficients of the Spatial Filter boxes

Post by hackr »

Thank you, Jussi!

Post Reply